Efficient zone identification approach for the recognition of online handwritten Gurmukhi script

Abstract

In this paper, a real-time recognition approach for online handwritten Gurmukhi character combinations with matras (Gurmukhi Vowels) has been addressed. Moreover, various challenges in the recognition of online handwritten Gurmukhi script have also been discussed. The strokes for writing Gurmukhi words can be drawn in one of the three horizontal zones, namely upper, middle and lower. With the huge variation in handwriting styles of writers, zone identification of online capturing strokes has become one of the major issues in Gurmukhi script recognition. In this connection, a robust zone identification algorithm has been proposed in this paper. We have considered 93 stroke classes (12 for upper zone and 81 for lower zone) to implement the proposed zone identification algorithm. The statistical tool, support vector machine, has been employed as the classifier for stroke classification. A total of 52,500 word samples were collected from 175 writers in order to train the classifier. The proposed zone identification algorithm yielded an accuracy of 99.75% when tested on a data set of 21,500 character combinations with matras, written by 10 new writers.